education · Updated 2026-02-23

Prompt Engineering Basics

A practical framework for writing prompts that reliably produce usable outputs.

Prompting is not magic wording. It is specification: you reduce ambiguity and increase constraints so the model has less room to guess.

The 5-part prompt formula

  1. Role: who the assistant should behave as (copywriter, recruiter, QA lead).
  2. Objective: what success looks like (increase demo requests, reduce churn).
  3. Context: audience, product, industry, situation, examples.
  4. Constraints: length, tone, structure, must-include, must-avoid.
  5. Verification: assumptions, risks, and a checklist.

Good constraints that improve output quality

  • Format: “Use H2 sections + bullets + a final checklist.”
  • Length: “120–160 words” or “3 bullets max.”
  • Audience language: “8th grade reading level” or “industry jargon OK.”
  • Grounding: “If you don’t know, say so and ask questions.”

Example prompt

Act as a conversion copywriter. Objective: write a landing page hero section that increases demo requests. Context: B2B scheduling software for small clinics. Audience: office managers. Constraints: 1 headline (<= 9 words), 1 subhead (<= 22 words), 3 bullets, 1 CTA. Verification: list 3 assumptions and 3 risks.

Iteration workflow

  1. Generate v1 quickly.
  2. Pick the best parts (headline, angle, structure).
  3. Re-run with stronger constraints and “keep these parts.”
  4. Finalize and verify claims.

Next: Common Prompt Mistakes and Verification Checklist.

FAQ

What is a good prompt structure?
A good prompt includes a role, objective, context, constraints, and a verification step (assumptions + checklist).
How long should a prompt be?
As long as needed to remove ambiguity. Short prompts can work if the task is simple; complex tasks benefit from explicit constraints and examples.
How do I stop generic outputs?
Add audience specifics, objections, examples, and formatting constraints. Also ask for two variants and a critique pass.
Should I ask the model to cite sources?
For factual claims, yes—ask for sources or a list of statements that require verification, then confirm independently.